Skip to main content
AI-BIM App includes an integrated AI assistant powered by OpenAI that can answer questions about your loaded IFC models.

​
How AI Queries Work

The AI system analyzes your IFC file data and responds to natural language questions:
  1. When you load an IFC file, the system extracts relevant data
  2. Property information is organized and prepared for AI analysis
  3. You ask questions in plain English
  4. The AI analyzes the extracted data and provides concise answers
Reference: ChatGpt component at /src/bim-components/ChatGpt/index.ts:30

​
Accessing the AI Panel

The AI query interface is available in the Load to tab:
1

Load an IFC File

First, ensure you have loaded an IFC model. The AI needs model data to analyze.
2

Navigate to Load to Tab

Click on the Load to tab in the floating toolbar at the center of the viewport.
3

Find the AI Panel

The AI Panel section contains a text input field labeled “Write Question” and a response area.
Reference: AI Panel UI at /src/bim-components/ChatGpt/src/user-interface.ts:42

​
Making Your First Query

1

Type Your Question

In the “Write Question” text input field, type your question about the model. For example: “How many walls are in this building?”
2

Click Query Button

Click the Querry button to submit your question to the AI.
3

Wait for Response

The AI processes your question against the model data. This may take a few seconds.
4

Read the Answer

The AI’s response appears in the text area below the input field. Responses are concise and based only on the available data.
5

Ask Follow-up Questions

Click Reset to clear the previous query and response, then ask another question.
Reference: Query submission at /src/bim-components/ChatGpt/src/user-interface.ts:14

​
What Data is Available to Query?

The AI has access to specific IFC data extracted from your model:

​
Material Information

  • Material names and types
  • Material associations with elements
  • IFCMATERIAL entities
Reference: Material extraction at /src/bim-components/LoadIfc/index.ts:94

​
Building Storey Data

  • Building level names (Ground Floor, Level 1, etc.)
  • Elements contained in each storey
  • IFCBUILDINGSTOREY entities
  • Spatial containment relationships

​
Element Properties

  • Element types (walls, slabs, beams)
  • Property sets (IFCPROPERTYSET)
  • Element quantities (IFCELEMENTQUANTITY)
  • Defined properties for each element
Reference: Property extraction at /src/bim-components/LoadIfc/index.ts:47

​
Entity Types

The system extracts data for:
  • IFCWALL and IFCWALLSTANDARDCASE
  • IFCSLAB
  • IFCBEAM
  • IfcRelAssociatesMaterial
  • IfcRelContainedInSpatialStructure
Reference: Valid entities filter at /src/bim-components/ChatGpt/index.ts:39

​
Example Questions

​
Material Queries

How many different materials are used in this building?
What materials are used for the walls?
List all the materials in this model.
What is the most common material?

​
Storey Information

How many building storeys are in this model?
What are the names of each floor?
How many elements are on the ground floor?
Which level has the most elements?

​
Element Counts

How many walls are in the building?
How many slabs and beams are there?
What types of elements are in this model?
Count the total number of structural elements.

​
Property Queries

What properties are defined for wall elements?
Are there any element quantities specified?
What property sets are available?

​
Crafting Effective Questions

​
Be Specific

Instead of “Tell me about this building,” ask “How many walls are on Level 1?”

​
Reference IFC Terms

The AI understands IFC terminology. Use terms like “IFCWALL”, “building storey”, “property sets”, and “material associations”.

​
Ask One Thing at a Time

Break complex questions into simpler ones: “How many walls?” followed by “How many slabs?”

​
Request Counts and Lists

Questions about counts and lists work best: “Count the materials” or “List all building storeys”.

​
Understanding AI Responses

The AI system is configured to provide concise, fact-based answers:

​
Response Characteristics

  • Concise: Answers use as few words as possible
  • Data-driven: Only information from the IFC file is included
  • No assumptions: If data isn’t available, the AI won’t guess
  • JSON-based: The AI analyzes structured JSON data extracted from IFC
Reference: System prompt at /src/bim-components/ChatGpt/index.ts:78

​
When the AI Says “Not Found”

If the AI responds that information isn’t available:
  • The IFC file may not contain that specific data
  • Property sets might be incomplete
  • The data may be in a format not currently extracted

​
AI System Architecture

Understanding how the system works helps you ask better questions:

​
Data Extraction Process

1

IFC File Loaded

When you load an IFC file, it’s read as text and converted to an array buffer.
2

Properties Extracted

The system extracts properties for specific entity types using the IFC Relations Indexer.
3

JSON Conversion

Extracted properties are converted to JSON strings for AI analysis.
4

Filtered Data

Only relevant IFC entities are included to reduce data size and improve response speed.
Reference: File loading at /src/bim-components/LoadIfc/index.ts:63

​
Query Processing

1

User Submits Question

Your question is sent to the OpenAI API along with the extracted IFC data.
2

AI Model: GPT-3.5-Turbo

The system uses GPT-3.5-Turbo for fast, cost-effective responses.
3

System Prompt Applied

A system prompt instructs the AI to answer only based on provided data and to be concise.
4

Response Returned

The AI’s answer is displayed in the text area.
Reference: GPT query processing at /src/bim-components/ChatGpt/index.ts:67

​
Data Filtering for Performance

The system includes an optional data filtering mechanism:
  • Filters IFC file content to include only relevant entities
  • Reduces data sent to the AI API
  • Improves response time and reduces API costs
  • Focuses on: materials, walls, slabs, beams, storeys, and relationships
Reference: Data modification at /src/bim-components/ChatGpt/index.ts:37
The filtering function is implemented but may not be active by default. Check the code configuration if responses seem incomplete.

​
Limitations and Considerations

​
Data Availability

  • AI can only answer questions about data present in the IFC file
  • Some IFC files have minimal property sets
  • Geometric data (shapes, dimensions) may not be fully queryable

​
Response Time

  • Queries typically take 2-5 seconds
  • Large IFC files with extensive data may take longer
  • API response times vary based on OpenAI service load

​
API Costs

  • Each query consumes OpenAI API credits
  • Consider data filtering for cost optimization
  • GPT-3.5-Turbo is chosen for cost-effectiveness

​
Privacy

  • IFC data is sent to OpenAI’s servers
  • Ensure compliance with your data privacy requirements
  • Sensitive project data should be handled according to your security policies

​
Troubleshooting

​
”No file data, cant answear” Alert

  • Ensure you’ve loaded an IFC file before querying
  • Reload the file if necessary
  • Check that the file loaded successfully (visible in viewport)
Reference: File data check at /src/bim-components/ChatGpt/src/user-interface.ts:19

​
Empty or Generic Responses

  • Your IFC file may lack the requested data
  • Try more specific questions using IFC terminology
  • Check that property sets are defined in the source BIM model

​
Query Button Doesn’t Respond

  • Ensure you’ve entered a question in the text input
  • Check browser console for error messages
  • Verify your OpenAI API key is configured (developer setting)

​
Responses are Incomplete

  • The AI is configured for concise answers
  • Ask follow-up questions for more detail
  • Try rephrasing your question to be more specific

​
Tips for Best Results

Load IFC files that include comprehensive property sets for the most detailed AI responses.
Start with simple counting questions to understand what data is available, then ask more complex queries.
Use the Reset button between unrelated questions to clear the previous context.
If a question doesn’t work, try using exact IFC entity type names like “IFCWALL” instead of just “wall”.

​
Next Steps

Combine AI queries with other features:
  • Use data visualization to see element distributions
  • Apply measurements to verify dimensions mentioned in AI responses
  • Cross-reference AI answers with the element selection panel for detailed properties

Build docs developers (and LLMs) love

Get started for freeTalk to us